GPT-SoVITS
App : A product you use directly
Voice cloning and synthesis from a few seconds of audio, running locally.
The most popular local voice cloning tool; responsibility for the cloned voice is yours.
Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.
What it does, concretely
Reproducing a voice from a short recording used to require hours of data and significant resources. GPT-SoVITS makes it possible to clone a voice from as little as 5 seconds of audio, or to fine-tune a model with one minute of recording for a more accurate result.
The tool runs entirely locally through a web interface: you provide a reference audio sample, enter the text to synthesize, and the model generates speech in the target voice. An optional fine-tuning step improves similarity by slicing, transcribing, and annotating training data directly within the interface.
GPT-SoVITS is aimed at content creators, voice actors, developers, and researchers who want to experiment with voice synthesis without sending their data to a third-party service. Synthesis is supported in English, Japanese, Korean, Cantonese, and Chinese, including in a language different from the reference sample.
However, installation requires Python, Conda, and basic command-line knowledge. On Mac, training quality is reduced because the Apple Silicon GPU is not yet fully supported for that step. This tool is not suited for general audiences without technical guidance.
What it offers
- ·Zero-shot voice cloning: generates speech in a target voice from a sample as short as 5 seconds.
- ·Few-shot fine-tuning: improves voice similarity by training the model on approximately one minute of annotated audio.
- ·Multilingual synthesis: produces speech in English, Japanese, Korean, Cantonese, and Chinese, even when the reference voice is in a different language.
- ·Integrated voice/accompaniment separation: isolates a voice from a musical recording before using it as a reference.
- ·Automatic transcription (ASR): segments and transcribes training data using Fun-ASR-Nano, SenseVoice, or classic FunASR.
- ·Local web interface: all steps (preparation, training, inference) are accessible from a browser with no internet connection.
- ·Docker deployment available: an official Docker image allows running the tool in an isolated environment on Linux.
In practice
- Price
- Free
- Commercial use
- Commercial use: yes
- Available on
- windows, linux, macos
- Licence
- MIT
- Runs locally
- Yes, 4 GB VRAM recommended
- FR interface
- No
Security and data
- Rating
- Security: one caveat, see the card
- Your data
- Data stays local
- Known incidents
- Utilisable pour des deepfakes vocaux ; marquage exigé par l'article 50 de l'AI Act depuis août 2026.
- Maintenance
- Active
- Last release
- 6 June 2025
Links and repository checked on 25 October 2026 · added on 1 January 2024
How to install GPT-SoVITS
conda create -n GPTSoVits python=3.10 && conda activate GPTSoVits && bash install.sh --device CPU --source HF